Song similarity is a complex, ill-defined problem. We don't currently provide an API for it.
On a small number of tracks, you could define the similarity metric yourself by looking at the timbre, pitch, rhythm, tempo, and subjective attribute data. The summary of that data and the combination of what's important and when, is what makes it particularly difficult.
Octave errors (doubling or halving) with tempo can happen. People also sometimes disagree on the best answer. If you send us a list of TrackIDs for those tracks, we can have a look at what might be going on there.